Updated September 26, 2026 · 6 min read
GPT6: OpenAI's GPT-6 family, explained
"GPT6" (also written GPT-6) is the name of OpenAI's 2026 model generation. It launched on September 3, 2026 with the flagship GPT-6 Astra, and was expanded on September 22, 2026 with two cheaper siblings: GPT-6 Sol and GPT-6 Luna. Here's what each model does and what it costs.
The three GPT6 models
OpenAI split the GPT-6 generation into three tiers instead of one model for everything:
- GPT-6 Astra — the flagship. Released September 3, 2026. It's the best model OpenAI has shipped, built for autonomous computer use: browsing, filling forms, running multi-step workflows, and producing polished documents and slides. When people say "GPT-6", they usually mean Astra. Deep dive: what is GPT-6 Astra?
- GPT-6 Sol — the mid-tier workhorse, released September 22, 2026. Built for complex coding and agentic workflows at a much lower price. It's the successor to GPT-5.6 Sol.
- GPT-6 Luna — the efficiency tier, also September 22, 2026. Made for high-volume work: summarizing, extraction, classification, routing. It's also the first GPT-6 model that Free and Go subscribers can use, through the ChatGPT desktop app.
One naming note: in ChatGPT's regular chat on Pro plans, Astra shows up as "GPT-6 Pro". "GPT6" and "GPT-6" mean the same thing — the hyphen is just styling.
GPT6 pricing (API, per million tokens)
| Model | Input | Output | Cached input | Released |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | $1.00 | Sep 3, 2026 |
| GPT-6 Sol | $2.00 | $10.00 | $0.20 | Sep 22, 2026 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 | Sep 22, 2026 |
Full consumer plan breakdown here: GPT-6 Astra price. Prompts over 272K input tokens are billed at 2× input and 1.5× output rates. Batch and Flex modes take another 50% off.
Shared specs
- Context window: 1,050,000 tokens (about 922K usable for input) across all three models
- Maximum output: 128,000 tokens
- Input: text and images · Output: text
- Reasoning effort settings: low through maximum (Luna adds a "none" setting)
- Astra's knowledge cutoff: April 30, 2026
- All three support streaming, function calling, structured outputs, and computer use
GPT6 benchmarks (Astra, OpenAI-reported)
| Benchmark | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|
| OSWorld 2.0 (computer use) | 72.6% | 65.7% |
| FrontierMath Tier 4 | 97.6% | 83.0% |
| ARC-AGI-3 | 99.9% | — |
| Terminal-Bench 4.0 | 57.9% | 37.3% |
| GPQA Diamond | 96.0% | 94.6% |
| BenchCAD (3D reconstruction) | 95.9% | 83.3% |
In the OSWorld latency simulation, Astra finished tasks in roughly 40 minutes versus 75 for GPT-5.6 Sol — about 47% faster. More detail: full benchmark list.
How to get GPT6
- Free / Go ($8): GPT-6 Luna through the ChatGPT desktop app; no Astra access.
- Plus ($20): limited Astra via the Work screen and Codex.
- Pro ($100 / $200): full Astra access, labelled "GPT-6 Pro" in chat. See free & plan access.
- Developers: API via OpenAI, Microsoft Foundry (Azure) and AWS Bedrock.